By Verlon Johnson, Chief Government and Corporate Affairs Officer, EVP
Last summer, a tornado came through my neighborhood, leaving trees and branches scattered around homes, and damage, disruption, cleanup, and a lot of work ahead. In the midst of it, I saw a dog roaming the neighborhood, looking at the same scene and instead seeing the greatest collection of chew toys that he had ever been given.
Same situation. Different perspective.
I shared this story in a Safe AI in Medicaid Alliance (SAMA) session at MESC 2026, and I find myself reflecting on it as I look forward to our next meeting on November 18. The story captures something important about where we are with artificial intelligence in Medicaid.
AI is disruptive. It is moving quickly, and there are legitimate questions about privacy, transparency, bias, accountability, data, workforce, governance, and what decisions we should ever allow a machine to make. We should not minimize those risks. But if all we see is disruption, we can miss the opportunity. And if all we see is opportunity, we can miss some very real risks.
Our responsibility as Medicaid leaders is to see both.
We’ve seen the conversation has also changed significantly over the past year. We are no longer talking primarily about what AI might someday mean for Medicaid. AI and automation are already being used across healthcare and Medicaid. The more important questions now are where these tools can create real value, where they should not be used, what requires human judgment, and what safeguards need to be in place.
The goal isn’t more AI. The goal is better Medicaid.
That is exactly why SAMA exists.
States should not have to answer the same questions in 50 different ways. And technology companies should not be developing solutions for Medicaid without the people who understand the program, its beneficiaries, and its operational realities at the table.
SAMA brings those perspectives together. State Medicaid leaders, policy experts, technologists, researchers, and implementation partners are working side by side to move AI from policy and possibility to safe, responsible Medicaid practice.
What excites me most, as an Acentra Health contributor to SAMA, is the breadth of knowledge and experience at the table. We have people who understand Medicaid policy and operations, people building and implementing technology, and people thinking deeply about governance, data, risk, and beneficiary experience. Bringing those perspectives together allows us to challenge assumptions, learn from one another, and solve problems in ways none of us could do alone.
And one principle has remained central to that work. As Sean Harrison, Acentra Health's Chief AI Officer and one of the leaders behind SAMA, often reminds me, the goal is to keep humans in the lead. AI can help, but people remain responsible for governance, validation, judgment, and accountability.
Over the past year, Daniel Hallenbeck from Optum has helped SAMA turn those principles into practical resources, including a shared Safe AI definition, a Medicaid AI use-case and risk-evaluation framework, implementation guidance, a model state Medicaid AI policy, and work aligned with the NIST AI Risk Management Framework.
At MESC 2026, SAMA contributors demonstrated what that looks like in practice.
Rural Health Transformation: From Use Cases to Readiness
Julie Boughn of JBoughn Consulting and former Director of the Data and Systems Group at the Center for Medicaid & CHIP Services within the Centers for Medicare & Medicaid Services (CMS) introduced SAMA’s Rural Health Transformation work. The effort created a structure for examining where AI-enabled capabilities could support rural health priorities and which applications were practical enough to move into deeper development.
SAMA reviewed state Rural Health Transformation Program submissions, identified AI-related proposals, and used participant voting to select priorities for further work. Those priorities included continuous provider risk monitoring, real-time intelligent dashboards, and intelligent member outreach use cases that address distinct Medicaid needs while sharing common requirements around governance, data quality, transparency, and human oversight.
The Continuous Provider Risk Monitoring workstream, presented by Michael Worthington of Acentra Health and Pooja Yelsangikar of FEI Systems, is a good example of why state participation matters. The team initially envisioned a prescriptive playbook. State feedback changed the direction. States told the workgroup they were not short on AI use cases; they needed practical tools to help evaluate, govern, and defend AI-assisted uses before implementation. The work resulted in a readiness guide with a self-assessment, governance minimums, vendor and product questions, an evidence checklist, and a rural-access impact review.
The work also established an important boundary. AI can help surface patterns associated with waste and abuse, generate leads, and stratify provider risk, but it cannot independently establish fraud or replace the evidence and human judgment required for adverse action. That is an important distinction for Medicaid. Automation can help us identify where to look. People remain accountable for what happens next.
Brett Brown of AWS Public Sector presented the Real-Time Intelligent Dashboards workstream, which focused on another challenge consistently raised by states: trust. The team developed a five-layer framework spanning data sources and ingestion, data quality and governance, AI and machine-learning processing, dashboard and user experience, and continuous improvement. SAMA principles such as human oversight, transparency and explainability, privacy, bias awareness, and incremental trust were mapped across the framework. The resulting demonstration showed six ways AI-supported dashboards can help users explore information while retaining visibility into source data and validation.
The Intelligent Member Outreach workstream, presented by Sean Harrison of Acentra Health and Maria Perrin of Muze Health, focused on how AI might help states communicate with members more effectively. The team identified use cases ranging from education and policy simplification to personalized outreach, contact-center support, and assistance at stressful points in the member journey. But the guardrails are just as important as the use cases. AI can explain approved information, simplify communications, provide reminders, support navigation, and help staff retrieve information. It should not independently determine eligibility or noncompliance, approve or deny exemptions, reject documentation, issue adverse actions, resolve appeals or grievances, or prevent access to human assistance.
Across all three workstreams, there was a consistent message: innovation and protection are not competing objectives. Medicaid must do both.
SAMA contributors also emphasized the importance of transparency. You cannot govern what you cannot see. States need to understand where AI is being used, what data it relies on, how outputs are validated, how performance is monitored, when human review is required, and who is accountable when something goes wrong. Those are not questions to answer after deployment. They belong at the beginning of the conversation, including in procurement, contracting, governance, and implementation.
Human oversight is especially important when technology can affect a beneficiary's access to care or coverage. One way to describe it is this: automation can help us say yes faster. We need to be much more careful about allowing it to say no.
And underneath all of this is data readiness. AI cannot compensate for inaccessible, fragmented, unreliable, or poorly governed data. In some cases, it can amplify those problems. Responsible AI adoption therefore must be connected to the broader work states are already doing around modernization, interoperability, data quality, and system agility.
What was most encouraging at MESC was not simply how much SAMA has produced in a relatively short period of time; it was how the work evolved. States provided feedback when something was not useful enough. Workgroups changed direction. Teams moved from broad concepts to practical tools. And people from organizations that might ordinarily sit on different sides of a procurement or policy discussion worked together on shared Medicaid problems.
That is exactly how we believe this work should develop as SAMA contributors. We will not get every answer right the first time, and SAMA’s collaborative structure gives us a way to learn and adjust. AI will continue to change, and so will the federal and state policy environment. Medicaid cannot afford to sit on the sidelines until every question has been answered. Nor should we race toward technology simply because it is available.
There is a responsible path between those two extremes, and together SAMA’s members have an opportunity to build it. SAMA can be a place where states do not have to solve the same problem independently, where industry hears directly from states about what they actually need, where we share both successes and cautionary lessons, and where we ask difficult questions before technology is deployed rather than after something goes wrong.
SAMA's next phase should continue turning what we are learning into state-ready tools that can be tested, challenged, and improved in real Medicaid environments. That feedback loop is essential. Safe AI cannot be a static checklist; it must evolve with the technology, the evidence, and the needs of the people Medicaid serves.
SAMA remains open to states and partners interested in contributing to this work, sharing lessons, and learning from peers. The next meeting is November 18. Join the conversation.
Because ultimately, the goal isn't more AI. It's better Medicaid.